Hi, I'm Tsuki 👋

Into AI agents and the infrastructure underneath them — TypeScript / Python, open source.

🔨 Selected open-source contributions

• agentconnect — a multi-agent platform bridging IM to AI coding agents. Hardened the sandbox shim's Git exec policy: replaced the blanket git config admission with an exact managed-credential allowlist (#2894, merged).
• evalscope — a streamlined framework for efficient LLM/VLM evaluation and benchmarking. Fixed Terminal-Bench Codex agent compatibility: skipped the unsupported max_turns parameter for Harbor 0.24.0's Codex schema (#1846, merged).
• cherry-studio — an AI desktop assistant built with Electron and React. Fixed the translate language selector leaking raw language codes into its closed trigger: the display label now uses the flag and localized name, with the code kept for search filtering (#21993, merged).
• KiroCrew — a self-improving persistent workspace for AI coding agents (Python). Fixed prepare-pr's --reviewers pin silently disabling the unanswered CONCERNS gate: the pin-independent whole-design verdicts are now merged into the gate, so a design CONCERNS still blocks auto-merge (#21808, merged).
• EvoScientist — a self-evolving AI scientist harness for autonomous research (Python). Fixed routed LLM providers silently falling back to the default vendor key and sending the wrong API key to third-party endpoints: they now fail fast with a clear error naming the missing provider key (#499, merged).
• tabularis — an open-source desktop SQL workspace (Tauri + TypeScript) with built-in drivers and shipped driver plugins. Added clipboard-driven grid filtering: new = clipboard and IN (clipboard) cell actions build the clause through DriverCapabilities, read the clipboard via the platform adapter so the browser build works too, and parse pasted lists with de-duplication and a 1,000-value cap (#1112, merged).

📊 GitHub Stats